feat(engine): surface abandoned-earcut-worker count for leak observability

[Kohei-Wada]

Summary

The engine keeps no metric for how many earcut workers it has
abandoned. Each abandoned worker is a leaked OS thread (#305) that
burns a core forever; today the only way to notice the engine is
leaking is to watch process CPU / thread count from outside. Surface
the count so the leak is observable in-app.

Proposal

• Engine: EarcutWorker counts each TimedOut abandonment. The count
  accumulates naturally — Canvas (which owns the worker) is a
  persistent Renderer field, cleared per frame, not recreated.
  Expose leaked_count().
• Reporting path: carry the count from the render thread to the parent
  (extend the FrameSink payload or add a dedicated EngineEvent),
  then to the App.
• UI: surface it (status readout and/or a notify once it crosses a
  small threshold) so the user knows when a manual RestartEngine is
  worthwhile.

Acceptance

• Engine increments a counter on each earcut timeout (unit test driving
  a timeout triangulation).
• The count reaches the App and is visible to the user.
• Restarting the engine resets the observed count to 0.

Out of scope

• Automatic restart when the count crosses a threshold (separate
  follow-up; this issue is observability only).
• Fixing the underlying earcut hang (earcut worker leak: abandoned triangulators keep burning CPU after timeout #305 root cause: cancellable
  triangulator / pre-validation).
• OS-level thread/CPU monitoring — the engine-internal count is the
  precise, portable signal.

Relationship to other issues

• earcut worker leak: abandoned triangulators keep burning CPU after timeout #305 — the leak this makes observable. Observability, not the fix.
• Pairs with the manual RestartEngine command (feat(app): restart-engine command — recycle the engine subprocess #383, mitigates earcut worker leak: abandoned triangulators keep burning CPU after timeout #305):
  this provides the signal for when to invoke it, and is the intended
  trigger source for a future runaway auto-restart.